The Scripted Loopholes are a series of media art works that investigate different aspects of computational propaganda.

The Advertiser works with the Facebook Ads Interests, a list that Facebook autonomously creates for every profile. It is partly based on direct user behavior (liking, commenting), and partly predicted by a black box algorithm. The list is categorized into hobbies, traveling, entertainment and more. Facebook’s prediction of my interests was sometimes true, but mostly wrong. It supported the hypothesis that algorithms are good at collecting but bad at contextualizing. Nevertheless, it is possible to make assumptions about my attitude, motivations and behavior from this list. A large number of entries where unknown to me, therefore the work includes information from Wikipedia for each term. Furthermore, the tool compares every advertisement interest with things I've liked in the Network to understand for which entries I'm responsible and which ones Facebook creates autonomously.

The Messenger was inspired by the uncovering of Facebook selling private chat records to companies like Netflix and Amazon. I was curious how much information an algorithm could derive from a chat between a friend of mine and me. The Google-developed Cloud Natural Language API served as a reference algorithm for the analysis. It provides different types of unstructured text analysis including sentiment, entity, and syntax analysis as well as content classification. Inspired by the Facebook Ads Interests list discovered in The Advertiser, I chose to run an entity analysis on my chat to identify dates, persons, contact information, organizations, locations, events, products, and media types. Although the results contain a lot of noise and wrong predictions, it is still possible to assume the personal address, profession, and country of origin of my friend from the results (Hint: he moved in the meanwhile).

The Network visualizes political activity on Twitter during the Swiss votes of October 2019. Between the 17th and 20th of October, I collected around 12’000 Tweets with the help of a bot-like script written in Nodejs. The collected Tweets were pruned and geographically mapped across Switzerland. While the complete data set amounts to ruffly 12'000 Tweets, the work at present only visualizes 3456 Tweets, because for many the geographical origin could not be determined beyond doubt.